Multiple Object Tracking in aerial vehicle overhead video

This paper proposes a multiple Object Tracking algorithm for drone overhead video, which solves some of the specific problems in this field. By studying the characteristics of small and dense targets in the UAV overhead shooting video, using the self-supervision technology to innovate the dynamic mask structure, combined with the existing multiple Object Tracking idea of first detection and then tracking, we designed our multiple Object Tracking algorithm, and finally trained and tested on the Visdrone dataset, which got good results and proved the superiority of our algorithm in the UAV overhead video.